Currently, the geometry aware sparse grids allow us to use only simple stencils for image classification on a normal resolution. In this thesis, I present the data hierarchy, a set of different coarsened images of the original data, that allow us to apply complex stencils on lower resolutions while still being able to process the original image through simple stencils. By exploiting the data hierarchy the accuracy in some cases is improved by nearly 5%. I also present the class of hierarchical parent stencils that can work vertically on the data hierarchy and uses fewer interaction terms allowing us to establish a data hierarchy without necessarily increasing the number of grid points. Smart use of the data hierarchy and multiple stencils allows us to apply sparse grid image classification in cases that would have been infeasible before.     «

Currently, the geometry aware sparse grids allow us to use only simple stencils for image classification on a normal resolution. In this thesis, I present the data hierarchy, a set of different coarsened images of the original data, that allow us to apply complex stencils on lower resolutions while still being able to process the original image through simple stencils. By exploiting the data hierarchy the accuracy in some cases is improved by nearly 5%. I also present the class of hierarchical...     »